Content
70%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an unusually rigorous, well-sequenced instruction skill: concrete paths, defaults, decision order, explicit stop conditions, and a recheck loop make the workflow highly followable. Its weaknesses are token efficiency and structure — a generic interaction contract is duplicated between the intro and section 5, and all edge-case detail lives inline in a single 180-line file with no reference files.
Suggestions
Move the 'Runtime-aware questions' and 'Shared project skill-context' sections into a shared one-level-deep reference file, keeping a one-line pointer in SKILL.md, to remove ~30 lines of family-wide policy from this skill.
Deduplicate section 5's question-tool rules (plain-text fallback, cancellation-is-not-approval, non-interactive behavior) by referencing the interaction contract once instead of restating it.
Move section 2's inventory edge cases and localized heading-equivalent lists into a reference file, keeping the milestone-inventory steps themselves inline as the overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense rule-writing without concept explanations Claude already knows, but the generic interaction contract ("AskUserQuestion and Questions below describe the intended interaction...", "An empty/cancelled response... is not an answer or approval") spans ~21 lines of family-wide policy, and section 5 restates those same rules ("Cancellation, empty replies, preselected options and elapsed time are not approval"). This matches 'mostly efficient but includes some unnecessary explanation or could be tightened'; it is above 2 because there is little padding beyond the duplicated policy, and below 4 because the duplication and inline boilerplate are more than minor. | 3 / 5 |
Actionability | Guidance is largely executable for an instruction-only skill: exact paths (".ai-factory/config.yaml", "paths.plans"), concrete defaults ("fast=1, full=2, ultra=3 points"), exact size mapping ("S/M/L/XL = 1/2/3/4+ points"), and a copy-paste-ready output block ("/aif-plan full Add mail personalization helpers..."). It stays at 4 rather than 5 because some direction remains abstract ("a brief targeted code read", "explain the shared module/direct continuation") and only the planning-request portion of the output is exemplified, not a full recommendation with alternatives. | 4 / 5 |
Workflow Clarity | Sections 1–5 form a clear numbered sequence with explicit validation and feedback loops: ordered mode-resolution rules ("If present but invalid, warn and use ultra"), explicit stop conditions ("Missing roadmap → suggest /aif-roadmap and stop"), uncertain-by-default handling ("Ambiguous linkage or task sections leave claim/progress coverage uncertain, never unclaimed by default"), and a post-selection recheck ("After explicit selection, recheck readiness, claims, mode and source revisions"). This matches the 5 anchor's clear sequence, explicit validation steps, and error-recovery paths; nothing significant is missing for a read-only workflow. | 5 / 5 |
Progressive Disclosure | Sections are well organized with clear headers, but the ~180-line body has no bundle files at all, and content that plausibly belongs in separate files is inlined: the generic "Runtime-aware questions" and "Shared project skill-context" policy sections, plus detailed edge-case lists like the localized heading equivalents ("## Задачи", "## Связь с дорожной картой", "Веха:"). This fits 'some structure but could be better organized; content that should be separate is inline'; it is above 2 because structure and navigation are good, and below 4 because SKILL.md carries the full detail load rather than pointing to one-level-deep references. | 3 / 5 |
Total | 15 / 20 Passed |